您好,欢迎访问三七文档
当前位置:首页 > 行业资料 > 国内外标准规范 > PCA人脸识别实验报告
Assignment1:FaceRecognitionusingPCAIntroductiontoBiometricsXunHuang0827121TU/ex.huang@student.tue.nl2012/10/81.IntroductionThisassignmentaimstobuildasimplyconstructedrecognitionsystemusingthestandardPrincipalComponentAnalysis(PCA)methodtoidentifyfaceimageandnon-faceimage.Afterthat,thetestsetwillbeused,toanalyzetheperformanceofthePCAapproachforvaryingfeaturesizes.2.Initiate:ReadthetrainingsetWehavea‘FaceData’containingimagesof40persons,with10imageseachofthem.Andwhatweshoulddofirstistoextracttheinformationofthefirst5imagesofeachpersonastrainingset.AccordingtothePCAmethod,weshouldreshape2Dimagesinto1Dimagevectors.Inthisexperiment,weprefercolumnvectortorowvector.So,aR*200matrixwillbecreated,withR=56*46,whichisthewidthandheightofajpgimage.3.Plotthefirst20EigenfacesforthePCAapproach.Figure1:EigenfacesforstandardPCA4.ReconstructasampleimagefromthetestsetusingPCAfeaturesofgivensizesThearrayIchoseasPCAfeaturesizeis[2,5,10,20,40,60,100,150,200,400,1000,2000].ThesampleimagewhichIchoseisthe(12,10).SeeFigure2.Figure2:Reconstructasamplefaceimage(12,10).Thenextscreenshotisforreconstructinganon-faceimage.ItistheEiffelTower.Seefigure2-(b).Figure3:Reconstructanon-faceimage.Comment:IntheFigure2,wecanseethatfromthe100,therudimentofthisman’sheadportraitreveals,hishairstyleiscleartousfrom600.Let’sobservetheFigure3,wenoticethatthisimagelookslikeatowerafter1000,beforethe1000,itjustseemslikeanoseofaman.Sowecandrawaconclusionthatthefacedatabaseplaysbetteronidentifyingthehumanfaceimagethanidentifyingthenon-faceimage.5.AnalysisontheperformanceofPCAmethodThetotalvarianceexplainedbyselectingthebiggestKeigenvectorsiscomputedby∑∑.Figure4:computingtherank-1identificationaccuracyFigure5:totalvarianceexplainedNumberofEigenfacesVarianceExplainedRank-1RecognitionRate20.3328500.320080.6140530.8050100.6547910.8400150.7223450.8400200.7673500.8650270.8109860.8750500.8923730.8900900.9522780.89001200.9745250.90001500.9885570.90502001.0000000.90502501.0000000.90505001.0000000.905010001.0000000.905015001.0000000.905020001.0000000.9050Table1:TheperformanceofPCAbasedfacerecognizerwithrespecttofeaturedimensionality6.SummaryWecanseefromthetablethatwhentheNumberofEigenfacesreaches8,theRecognitionRateincreasesrapidlytoaround0.8,andkeepincreasinggraduallyinthewakeofNumberofEigenfaces.Andthenstopsincreasingat0.9050whennumberofEigenfacesis150.Andwhenthenumberofeigenfaceupto200,thevariancesexplainedreaches1,itmeansthatallthefollowingeigenvaluesisofnouseanymore.The1-200eigenvectorscontainsallthedataweneedtothisexperiment.
本文标题:PCA人脸识别实验报告
链接地址:https://www.777doc.com/doc-1992188 .html